DeepQuery

What is DeepQuery?

DeepQuery sits above the databases an organisation has already approved — Postgres, Snowflake, BigQuery, MongoDB — and lets someone ask a question without first learning where the answer lives. It plans the retrieval, runs it read-only through a single gateway, and returns the answer alongside the query it ran and the rows it read. The point is not that a question becomes easier to ask; it is that the answer becomes checkable. An analyst can open the underlying result, disagree with it, and say exactly which join was wrong.

Problems it solves

• Ask a question of company data without writing SQL
• Reason across several databases that were never designed to be joined
• Trace a number back to the rows that produced it
• Cut the backlog of one-off analytical queries
• Give non-analysts a safe, read-only way to ask
• Keep access governed centrally instead of by shared credentials

Key capabilities

Natural-Language Query

Ask a question in plain language and get an answer built from connected data, without hand-writing the query.

Cross-Source Intelligence

Reason across several approved systems at once, so an answer is not limited to whichever database someone happened to open.

Evidence Trail

Every figure carries the record it came from, so a reader can check the answer instead of trusting it.

Agentic Investigation

Specialist agents pursue a question through several controlled steps rather than answering in one pass.

Confidence Ceilings

Confidence is capped by the strength of the underlying basis, so a weak source cannot produce a strong-sounding claim.

How it is built

One gateway, not many connections

Every query runs through a single governed path, so access is granted and revoked in one place rather than per person per database.

Read-only by construction

The query path cannot write. A question can be wrong; it cannot be destructive.

Never guess a schema

When a column or relationship is ambiguous, DeepQuery asks rather than inventing a join that silently returns plausible nonsense.

The query is part of the answer

You see what was run. That is what makes it reviewable by someone who knows the data better than you do.

How it works

How the work is structured, and what each step's output rests on.

1. 1

   Connect approved sources

   Source

   Register databases once, with the credentials and row limits they should run under.

2. 2

   Understand the schema

   Observed

   Catalogue tables, columns and relationships so questions resolve against what actually exists.

3. 3

   Plan the retrieval

   AI-inferred

   Decide which sources answer the question and in what order.

4. 4

   Execute read-only

   Source

   Run the plan through the gateway, under the registered limits.

5. 5

   Assemble the answer

   Source

   Combine results and attach the query and rows behind each figure.

6. 6

   Explain the basis

   AI-inferred

   Show what was read, what was inferred and where confidence is capped.

Example use cases

Each example has its own page you can read and share.

Ask DeepQuery

Illustrative examples. Results depend on your own data.

• “Which customers increased spending by more than 25% this quarter?”
• “Compare revenue, refunds and support volume by region for the last two quarters.”
• “Which tables feed this KPI, and which one changed last week?”
• “Show accounts with rising usage but falling invoice value.”

Security, governance and human control

Where the software deliberately limits itself.

Questions

Can DeepQuery change my data?

No. The query path is read-only. It can be asked a wrong question; it cannot write, drop or alter anything.

What happens when a question is ambiguous?

It asks. Guessing a join is how a system returns a confident, wrong number, so ambiguity is surfaced rather than resolved silently.

Do I need to know SQL?

No. The generated query is shown, which is useful if you do know SQL and want to check the reasoning.

Which databases are supported?

PostgreSQL, Snowflake, BigQuery and MongoDB ship today. Only integrations that actually work are listed.